Enhanced facial expression recognition by age

Current studies of facial expression recognition (FER) pay little attention to the age effect on the performance of expression recognition. In this paper, we propose to enhance expression recognition by age. Specifically, we propose a three-node Bayesian network to incorporate age information as privileged information, which is only available during training. During training phase, a full probabilistic model is constructed to capture the joint probability among image features, age, and expression labels. During testing, the conditional probability of expression labels given image features is obtained by using the Bayesian rule and marginalizing over age. Experiments are conducted on two databases, i.e. the Lifespan and the FACES. Experimental results of the significant hypothesis test prove the age effect on expression recognition. Expression recognition experiments demonstrate that using age information as privileged information can construct a better expression classifier than using facial images alone.